Report on “Intelligent Music Software” Presentation Held on March 6, 2014
| March 11, 2014 | Posted by COMauthor under COMSOC, CS, EDCAS, General |
Comments off
|
Report on IEEE Foothill Computer Society chapter meeting held March 6, 2014 at California Baptist University, Riverside
Professor Robert Keller came to give an updated presentation on the topic of his project at Harvey Mudd College titled ”Intelligent Music Software”. By Professor Keller’s definition, “Intelligent Music Software” makes decisions that aid its user.
To understand how Prof Keller views this definition of “artificial Intelligence”, consider how he evaluates the recent IBM efforts to make chess-champion players from computers and computer “logic”. Deep Blue (1977) and Watson (2011) were chess champions, but “non-learning”. On the other hand, TD-Gammon, a backgammon game player (1994) used a neural network for “learning”; he would claim this as “intelligent”, at least in the mode of having “artificial intelligence”.
The main effort of Prof Keller’s talk was to describe a much more limited AI type program that he has developed at HMC. This is called “Impro-Visor”. It is designed to help aspiring student musicians learn how to improvise jazz compositions.
This effort started with developing a means to record and edit audio tracks (Audacity, FOSS). The aim was to go from the audio domain (mp3, wav, etc format) into the Musical Instrument Digital Interface (MIDI). What is needed for MIDI are the melody and the chord structure; not literal transcription of the musical piece. Prof Keller then directs his students to look at the chord structure. His is trying to have his students “learn” from working on software that “learns”. Can the student develop the habit to improvise and then to improve upon the original composition? (In response to a question, Prof Keller does acknowledge and understand the difficulty of evaluating(subjectively??) how well the software scores on “learning”.)
The software developed by Prof Keller and others has become quite sophisticated. Starting with the chord progression, advise for entering notes can be generated, and “scored in a Four Color Scheme”(black, blue, green, and the to-be-avoided red notes). As the musician progresses, more advice is given via the software: Licks as short vocabulary segments; changes to probabilistic grammar, and embedded Markov chains of suggested chords as grammar. Other related programs (Deep Belief Networks, Restricted Boltzmann Machines ) were briefly discussed.
On the question of whether or not professional jazz musicians have used his Impro-Visor software, Prof Keller acknowledged that none have. It may be that they are capable, with their trained hearing, of generating and evaluating improvisions to their compositions in real time right now. No need for any software.
Can these Improv-Visor “suggestions” turned into jazz music be distinguished from the original artist and composition? Prof Keller says that he can readily distinguish between a recording of saxophonist John Coltrane and an Impro-Visor version of a pseudo-like-John-Coltrane selection.
It was an interesting presentation to our IEEE Foothill Computer Society chapter. We will be asking Prof Keller to give us a future talk on his music education / artificial intelligence project. In the meanwhile, we will confine our listening to the likes of jazz musicians Dave Koz, Boney James, Marc Antoine and Brian Culbertson on HD radio during our work commute.